基于GA-RBF算法的采煤工作面瓦斯涌出量预测研究

发布时间：2018-08-15 19:20

【摘要】：我国煤炭产量居世界首位,煤矿安全事故也频频发生,伤亡人数仅排在因交通事故伤亡之后。随着开采深度地不断加深,生产能力地提高,地质条件也更加复杂化,煤矿安全工作面临着巨大的挑战。瓦斯事故又是煤矿生产过程中的主要不安全因素,如何能准确快速地预测出瓦斯涌出量,对于瓦斯防治措施的制定有着积极意义。本文从开采因素和自然因素两个方面分别分析煤层的瓦斯含量、埋深、瓦斯压力、大气压力、风量、产能等方面对瓦斯涌出量的影响,指出传统预测瓦斯涌出量方法的局限性,不能将瓦斯涌出量与各个影响因素之间复杂的非线性关系清楚地表述。RBF神经网络自身的容错性和自适应性以及较强的非线性函数逼近能力,则能很好地克服这些缺点。 RBF神经网络具有搜索全局最优解和最佳逼近能力,其拓扑结构、隐节点数目、中心位置、宽度和权值是决定整个网络性能的关键因素。遗传算法作为一种全局优化算法,具有强鲁棒性,适用于解决训练速度慢、易陷入局部极小值等缺点的网络结构,自适应调整交叉概率和变异概率,能够避免重复搜索,并提高搜索效率。本文提出采用遗传算法优化RBF神经网络中的隐节点数目、中心位置、宽度和权值,有效地弥补RBF神经网络的不足,最后利用Matlab软件编程实现GA-RBF神经网络模型,应用此模型分别对两个采煤工作面进行瓦斯涌出量预测,得到了令人满意结果。
[Abstract]:China's coal production ranks first in the world, coal mine safety accidents also occur frequently, the number of casualties only ranks behind traffic accidents. With the deepening of mining depth, the improvement of production capacity and the complication of geological conditions, the work of coal mine safety is facing great challenges. Gas accident is also the main unsafe factor in coal mine production. How to predict gas emission accurately and quickly is of positive significance to the formulation of gas prevention and control measures. This paper analyzes the influence of gas content, buried depth, gas pressure, atmospheric pressure, air volume and productivity on gas emission from two aspects of mining factors and natural factors, respectively, and points out the limitations of traditional methods for predicting gas emission. The complex nonlinear relationship between the amount of gas emission and the influence factors can not be clearly expressed. The fault tolerance and adaptability of RBF neural network and its strong nonlinear function approximation ability can not be clearly expressed. RBF neural network has the ability to search for global optimal solution and best approximation. Its topology, number of hidden nodes, center position, width and weight are the key factors to determine the network performance. As a global optimization algorithm, genetic algorithm (GA) has strong robustness, which is suitable for solving the problems of slow training speed, easy to fall into local minimum and other shortcomings of network structure, adaptively adjusts crossover probability and mutation probability, and can avoid repeated search. And improve the search efficiency. In this paper, genetic algorithm is proposed to optimize the number, center position, width and weight of hidden nodes in RBF neural network, which can effectively compensate for the deficiency of RBF neural network. Finally, the GA-RBF neural network model is realized by using Matlab software programming. The model is used to predict the gas emission in two coal mining faces, and satisfactory results are obtained.
【学位授予单位】：安徽理工大学
【学位级别】：硕士
【学位授予年份】：2013
【分类号】：TD712.5

【参考文献】